Teaching a robot duck to butler for my cat

A virtual butler for my cat

Using OpenAI's GPT-6 Sol, the author trained a virtual Microduck robot from Pollen Robotics and Hugging Face to walk around and identify bengal cats.

Hours instead of weeks

The author coded and trained the virtual robot in just a few hours.Previously, this kind of work might have taken weeks.

Berkeley researchers test Astra

When OpenAI released Astra in September, University of California, Berkeley professor Ken Goldberg and his students wondered if it could help with simulations needed to train robots.

A full simulation in one hour

Within an hour, Astra produced a full three D simulation with all parameters set: camera position, lighting, background, and materials.

“What blew us all away was that within an hour it had produced a full 3D simulation with all the parameters set, the position of the camera, the lighting, the background, all the materials.”

Teaching the duck to walk

The author used GPT-6 Sol, an agentic harness from Nous Research called Hermes, and an Nvidia DGX Spark to train the Microduck.

Open source framework for robots

Jim Fan at Nvidia, Ken Goldberg, and others open sourced an agentic framework for robots called Aspire, which allows a robot to discover and use new skills.

Painting the Golden Gate Bridge

At the Conference on Robot Learning in Austin, Texas, Jim Fan demonstrated Astra controlling a robot arm told to paint a picture of the Golden Gate Bridge.

Impact on manufacturing and hobbyists

Getting robots to work reliably is still extremely difficult, but the ease and speed of training with AI could have enormous impact on manufacturing and e-commerce.Industrial robotics firms have shown interest.